Line Correction for Robot Go-Forward using
Image Detection Approach
Shih-Chang Hsia
National Yunlin University of Science and Technology,
Douliou, Taiwan
Email: [email protected]
Wei-Siang Fu
National Yunlin University of Science and Technology,
Douliou, Taiwan
[email protected]
Abstract - In this paper, we present a line correction
method for robot moving control based on image
detection approach. By statistics of horizontal vector, the
line deviation can be estimated, and then to control the
speed of motor to trim the moving direction. For real-
time control requirement, we presented a fast estimation
algorithm based on binary successively approximate
approach, to reduce the computation complexity about
72.3%. This method is successfully implemented on a
real wheel robot system, which can perform line
correction in real-time using camera imaging processing. Keywords-robot, motion search, line correction, image
I .INTRODUCTION
The use of Robots will be more and more in the future.
There are many kinds of robots for various applications.
Generally, there are two types for robot classifications
according to motion features: one is moving-able robot,
such as huamn-like and wheel-based robot, and the other is
fixed on worktable, such as machine arm. In the former, the
applications will be to close to people living. In the latter,
most of applications are in industrial manufactures. The
moving-able robot can be interactive with humans, which
require intelligent functions by sensing environment. For
robot control, we need many kinds of sensors, such as (1)
camera : like human eyes, to detection environment by
image recognition; (2) microphone: like human ears, to
listen environment sound; (3) ultrasonic sensor: can detect
the distance between robot and obstacles.
In this study, we proposed a smart robot for house
patrol using wheel-based design. This robot can monitor
house condition and send the videos to remote devices. such
as smart cell phones. For cover wide monitoring area, the
robot uses wheels to move to anywhere. However, we face
a main problem: the robot cannot forward go as providing
the same speed to all motors of wheels. The robot may loss
control, sometimes turn right and sometimes turn left
dependent on various floors. Friction of wheels to floor is
different in practical, causing the motor driving power for
wheel is different and the moving direction of robot is not
predicted when commanding to robot. To solve this problem,
we proposed vision-based to correction the error of
deviation. When finding the robot moving direction is not
predicted, the correct algorithm can compute suitable
parameters and send to motor to control the speed of left and
right wheels.
The rest of this paper is organized as follows. The
proposed linear correction algorithm for robot is presented
in Section II. The implementation and experiments are
described in Section III. The conclusions are marked in
Section IV.
II. PROPOSED ALGORITHM
In this paper, we employ image processing approach to
localize the robot moving. First, the sum of absolute
difference (SAD) is computed along horizontal direction,
which can be given by
,|),(),(| _
1
0
1
0
1
N
i
N
j
tt jpiqfjpiqfXSAD (1)
, where ft and ft-n are the current and the previous nth
frame respectively, (q,p) is the position of macro block, and
N is the block size. Compute horizontal vector from -k to
+k , and to find a minimum SAD_X and record the current
vector, as shown in Fig. 1.
Figure 2 shows the processing flowchart of the
proposed line correction for robot moving. The camera is
located on the head of robot. The image is sampled and sent
to flat compute using USB interface. The frame is first
stored to memory. Next, we can compute SAD_X using (1)
by the current frame and the previous frame within a NxN
block. From full-searching -k to +k vectors, one can find a
minimum SAD_X that is the horizontal motion vector. If the
SAD_X is larger than the threshold th1, implying that the
robot may be not on the line. Then the SAD_X value and its
relative motion vector are saved to register. When a
positive motion vector is found, the data is accumulated to
Reg1, otherwise to Reg2. The next block is processed by
using the same flow. When all blocks had been processed
for one frame, we can find a maximum SAD_X value
between Reg1 and Reg 2. Then the motion vector from the
maximum SAD_X is averaged as a control parameter for
controlling motor of wheel.
International Conference on Computer Science and Service System (CSSS 2014)
© 2014. The authors - Published by Atlantis Press 545
Figure 2 The processing flow of line correction algorithm.
If the motion vector is positive, the deviation direction of
robot may turn right. Such a case, we speed-up the motor of
the right wheel and speed-down the motor of the left wheel.
Then checking the motion vector of next frame, we continue
to trim until to the all over motion vector is less than the
threshold. Oppositely, if the motion vector is negative, the
deviation direction of robot may turn left. Then, we speed-
up the motor of the left wheel and speed-down the motor of
the right wheel to force the robot to right for line correction.
For robot direction control, we need to control the
moving direction of robot immediately, so a real-time issue
is very important. From (1), if N=16, we require 256
subtractions and 255 additions for computing one SAD
value. When k=16, to find one motion vector of one block,
we require 4352 subtractions and 4335 additions. The
computational complexity is too for real-time robot control.
To reduce computational complexity, we present a fast
algorithm to find the horizontal vector for robot control
based on binary splitting approach. First, we compute the
SAD_X value of the points -k/2, 0, +k/2, and then to find
the minimum SAD_X among three points, as shown in Fig.
3(a). If the minimum SAD_X is located at the points -k/2,
then the central point is moved to -k/2. We compute
SAD_X of the points (-k/4+-k/2) and (-k/2+k/4), and to find
the minimum SAD_X among (-k/4+-k/2), -k/2, (-k/2+k/4)
points in the second step, as shown in Fig. 3(b). By using
Figure 3 (a) Compute SAD_X of -k/2, 0, +k/2 points in the
first step.
Figure 3 (b) Check the minimum SAD_X among (-k/4+-
k/2), -k/2, (-k/2+k/4) points in the second step.
Figure 3 (c) Find the horizontal vector in the last step.
546
this processing flow, we can find the final vector at the last
step, where the differential vector of last three points should
be one, as shown in Fig. 3(c).
For example, if the real vector is at -15, first (-8,0,+8)
points are checked in the first step. The minimum SAD_X
should be at the -8 vector. Then, the points of (-12, -8, -4)
are checked in the second step, where we only compute
SAD_X of -12 and -4 vectors and SAD_X of -8 had been
calculated in the first step. The minimum SAD_X would be
at the -12 point. Then the points of (-14, -12, -10) are
checked in the three step and find the minimum SAD_X at -
14. Finally, we calculate the SAD_X of -15 and -13 points,
and find the minimum SAD_X among -15,-14 and -13. The
differential vector is one that is confirmed at the last step,
and we can find the best vector at -15. Using this binary
successively approximate method, the computation can be
greatly reduced. In original full-search, we need to compute
SAD_X of 33 vectors when k=16. In this approach, we only
compute 3, 2, 2, 2 points in the first, second, three and last
step respectively. In total, only 9 points are checked. The
computation can save (1-9/33)=72.7%.
III. IMPLEMENTATION ON WHEEL ROBOT
The robot system is constructed by using two wheels that
is controlled by motor, and one all-direction wheel for prop
up the robot that is fixed at the tail. We built three ultrasonic
modules to calculate the distance from robot to obstacles.
Figure 4 shows the function block of the robot control
system. The control kernel used a micro-processor with a
single chip. The two external driver chips are used to drive
two motors to control two wheels for robot moving. A
lithium battery is used to supply the power of motor, and
using buck chip to drop the voltage for micro processor.
Two digital SR motor is used at the head of robot that can
be directly controlled for up/down and left/right for camera
using micro-processor. We used one blue-tooth module for
communicating with computer.
The flat computer is built in on the robot for
communication, computation and human interface. The
camera of head sent the video signals to flat computer, and
C-programming is used to develop the line correction
algorithm based on robot platform. Using the proposed fast
algorithm, we can process 3~5 frames to find the motion
vector. By statistics of horizontal vector, the deviation
direction can be estimated at once. The flat-computer sent
the estimated parameters to micro-processor through blue
tooth. According to the estimated parameters, the micro-
processor used PWM (pulse width modulation) to control
the speed of right-wheel and left-wheel motors and to
change the moving direction. When finding the moving
deviation is not on line, the micro-processor can command
the motor to correct it to line.
Figure 4 The robot control system.
Now we show practical results in experiments.
Figure 5(a) and (b) shown the results without line correction
at the 10 frame and 30 frame respectively, where two
driving motors used the same speed. Clearly, the robot
appears to turn right deviation. If not to correct, the robot
may turn in cycle.
(a)
(b)
Figure 5 The non-correct robot moving (a) at the 10 frame;
(b) at the 30 frame (turn right deviation).
547
Next, we used our proposed line correction method for robot.
The camera sampled the videos to computer processing and
to control motor immediately. Figure 6 (a) and (b) shown
the moving direction at the 10 frame and the 30 frame
respectively. From experiments, the result shown that the
robot can move on line after our line correction.
(a)
(b)
Figure 6 The proposed correct method for robot moving (a)
at the 10 frame; (b) at the 30 frame (direction correct).
V. CONCLUSIONS
In this paper, we study the line correction algorithm for
robot moving control. Using image processing approach, we
compute the over-all horizontal vectors to estimate the
deviation direction in statistics. To improve the control
speed for real-time purpose, the fast binary successively
approximate approach is proposed. The micro-processor can
change the motor speed 3~5 times per second. Experimental
results shown that the robot can move on line after our
correction processing. The method is very value for robot
navigation, patrol and auto localization system.
REFERENCES [1] K. Okuyama, T. Kawasaki, and V. Kroumov
"Localization and Position Correction for Mobile Robot
Using Artificial Visual Landmarks" IEEE International
Conference on Advanced Mechatronic Systems ,
August 11-13, 2011, pp. 414-418.
[2] A. Ryberg, M. Ericsson, A.-K. Christiansson, K.
Eriksson, J. Nilsson and M. Larsson, "Stereo Vision for
Path Correction in Off-Line Programmed Robot
Welding", IEEE Conf. Industrial Technology, 2010, pp.
1700-1705.
[3] D. Stavrou, D. G. Eliades, C. G. Panayiotou and M.
Polycarpou, " A Path Correction Module for Two-
Wheeled Service Robots Under Actuator Faults", 21st
Mediterranean Conference on Control & Automation
(MED), June, 2013, pp. 1119-1126.
[4] Peter Lim,Liyanage C. De Silva, Liu Jing, and Li Li
Ling,” Multimodal Approach to Human-Face Detection
and Tracking” IEEE Trans.INDUSTRIAL
ELECTRONICS, VOL. 55, NO. 3, ,PP. 1385 – 1393,
MARCH 2008.
[5] Liyuan Li, Shuicheng Yan, Xinguo Yu, Yeow Kee Tan,
and Haizhou Li, ”Robust Multiperson Detection and
Tracking for Mobile Service and Social Robots”, IEEE
Trans. SYSTEMS, MAN, AND CYBERNETICS—PART
B: CYBERNETICS, VOL. 42, NO. 5, PP. 1398 – 1412,
OCT. 2012.
548
